ThriveCart MCP Server
# ThriveCart MCP server for ChatGPT and Claude
ThriveCart has an API and webhooks, but no official MCP server was found, checked 20 November 2026. Cotlet.dev connects ChatGPT, Claude and other AI agents to your own signed-in ThriveCart account.
Source: [ThriveCart integration page](https://cotlet.dev/integrations/thrivecart-api-mcp-server?utm_source=github&utm_medium=readme&utm_campaign=thrivecart).
## What can your AI agent do?
The live page describes work across six checkout jobs and the ThriveCart screens your login can open:
- Find orders, check your refund policy, and prepare refunds that wait for your approval.
- Read product sales with bumps and upsells counted.
- Review, pause, upgrade or cancel subscriptions after confirmation.
- Review affiliate approvals, commissions and payouts.
- Filter and download transactions as CSV.
- Read or change products, checkout pages, funnels, coupons, bumps, upsells and downsells.
- Read reports for any period and read API and webhook settings.
Your agent cannot do more than your ThriveCart login. Payouts, payment processor settings and account settings stay in your own hands. Refunds, subscription changes, affiliate approvals and checkout changes wait for your OK.
## What can I ask?
> Find these customers' orders and prepare refunds after I approve them.
> What did each product sell last month, with bumps and upsells counted?
> Pause Leo's membership for two months at his request.
> Which affiliates are waiting for approval, and who referred the most sales this month?
> Download this month's transactions as a CSV for my bookkeeper.
> Make a 20% coupon for the Black Friday sale on the course bundle.
## Connect ChatGPT and Claude
1. [Get access to ThriveCart on Cotlet.dev](https://cotlet.dev/signup?app=thrivecart).
2. Enable ThriveCart in your Cotlet.dev console, open `https://thrivecart.com/` in your Cloud Browser, and sign in with your own business or team login.
3. In ChatGPT or Claude, add `https://cotlet.dev/mcp/v1/connect` as a custom MCP connector and allow access.
## Connect Codex and Claude Code
Create an API key in the Cotlet.dev console and set `COTLET_API_KEY` in your environment.
```toml
[mcp_servers.cotlet]
url = "https://cotlet.dev/mcp/v1"
bearer_token_env_var = "COTLET_API_KEY"
```
```bash
claude mcp add --transport http cotlet https://cotlet.dev/mcp/v1 --header "Authorization: Bearer $COTLET_API_KEY"
```
## Optional local MCP wrapper
Python 3.10+, standard library only. The repository provides seven generic browser helpers: `thrivecart_guide`, `thrivecart_status`, `thrivecart_open`, `thrivecart_read`, `thrivecart_click`, `thrivecart_type`, and `thrivecart_close`.
Set `COTLET_API_KEY` before calling the wrapper. Reads and writes still happen in your signed-in ThriveCart Cloud Browser; the agent asks before changes.
## Questions
**Does ThriveCart have an API or official MCP server?**
ThriveCart has an API and webhooks. No official ThriveCart MCP server was released or listed in the MCP Registry, checked 20 November 2026.
**Which ThriveCart account should I use?**
Use the owner login or a team login with the access you want the agent to have. The agent sees the screens and permissions this login can open.
**Is Cotlet.dev affiliated with ThriveCart?**
No. Cotlet.dev is independent and is not affiliated with ThriveCart.
## Access and control
Use your own ThriveCart account. You complete sign-in and any verification yourself. Disable ThriveCart for agents, reset the browser or delete the key in the Cotlet console to revoke access.
Cotlet.dev is independent and is not affiliated with ThriveCart.
TDQS
Scored across 7 tools
Most tools are clearly distinct: open/read/click/type/close map to distinct browser actions. thrivecart_guide and thrivecart_status are somewhat close as onboarding/readiness helpers, but the descriptions make their roles separable.
All tool names use a consistent thrivecart_ prefix followed by a short snake_case action verb. The pattern is predictable across the entire set.
Seven tools is well-scoped for a browser-automation server: setup, page reading, interaction, and cleanup are each represented without excess. Each tool appears to earn its place.
The surface covers the core browser lifecycle: guide, status check, open session, read page, click, type, and close. Minor gaps like explicit scrolling or waiting could be worked around via clicking/reading, but are not directly exposed.